anv: Implement VK_KHR_shader_atomic_int64

Reviewed-by: Lionel Landwerlin <[email protected]>
Reviewed-by: Caio Marcelo de Oliveira Filho <[email protected]>

4 files changed, 21 insertions, 2 deletions

diff --git a/src/intel/vulkan/anv_device.c b/src/intel/vulkan/anv_device.c
index de56926d935..d56e3593353 100644
--- a/src/intel/vulkan/anv_device.c
+++ b/src/intel/vulkan/anv_device.c
@@ -1051,6 +1051,14 @@ void anv_GetPhysicalDeviceFeatures2(
          break;
       }
 
+      case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_INT64_FEATURES_KHR: {
+         VkPhysicalDeviceShaderAtomicInt64FeaturesKHR *features = (void *)ext;
+         features->shaderBufferInt64Atomics =
+            pdevice->info.gen >= 9 && pdevice->use_softpin;
+         features->shaderSharedInt64Atomics = VK_FALSE;
+         break;
+      }
+
       case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETERS_FEATURES: {
          VkPhysicalDeviceShaderDrawParametersFeatures *features = (void *)ext;
          features->shaderDrawParameters = true;
diff --git a/src/intel/vulkan/anv_extensions.py b/src/intel/vulkan/anv_extensions.py
index 9d398f16cd3..d937e4e4524 100644
--- a/src/intel/vulkan/anv_extensions.py
+++ b/src/intel/vulkan/anv_extensions.py
@@ -104,6 +104,8 @@ EXTENSIONS = [
     Extension('VK_KHR_relaxed_block_layout',              1, True),
     Extension('VK_KHR_sampler_mirror_clamp_to_edge',      1, True),
     Extension('VK_KHR_sampler_ycbcr_conversion',          1, True),
+    Extension('VK_KHR_shader_atomic_int64',               1,
+              'device->info.gen >= 9 && device->use_softpin'),
     Extension('VK_KHR_shader_draw_parameters',            1, True),
     Extension('VK_KHR_shader_float16_int8',               1, 'device->info.gen >= 8'),
     Extension('VK_KHR_storage_buffer_storage_class',      1, True),
diff --git a/src/intel/vulkan/anv_nir_apply_pipeline_layout.c b/src/intel/vulkan/anv_nir_apply_pipeline_layout.c
index 356a56e47bb..9d25b9420fc 100644
--- a/src/intel/vulkan/anv_nir_apply_pipeline_layout.c
+++ b/src/intel/vulkan/anv_nir_apply_pipeline_layout.c
@@ -253,7 +253,7 @@ build_index_offset_for_deref(nir_deref_instr *deref,
 }
 
 static bool
-try_lower_direct_buffer_intrinsic(nir_intrinsic_instr *intrin,
+try_lower_direct_buffer_intrinsic(nir_intrinsic_instr *intrin, bool is_atomic,
                                   struct apply_pipeline_layout_state *state)
 {
    nir_builder *b = &state->builder;
@@ -262,6 +262,12 @@ try_lower_direct_buffer_intrinsic(nir_intrinsic_instr *intrin,
    if (deref->mode != nir_var_mem_ssbo)
       return false;
 
+   /* 64-bit atomics only support A64 messages so we can't lower them to the
+    * index+offset model.
+    */
+   if (is_atomic && nir_dest_bit_size(intrin->dest) == 64)
+      return false;
+
    if (!nir_deref_find_descriptor(deref, state))
       return false;
 
@@ -286,6 +292,8 @@ lower_direct_buffer_access(nir_function_impl *impl,
          switch (intrin->intrinsic) {
          case nir_intrinsic_load_deref:
          case nir_intrinsic_store_deref:
+            try_lower_direct_buffer_intrinsic(intrin, false, state);
+            break;
          case nir_intrinsic_deref_atomic_add:
          case nir_intrinsic_deref_atomic_imin:
          case nir_intrinsic_deref_atomic_umin:
@@ -299,7 +307,7 @@ lower_direct_buffer_access(nir_function_impl *impl,
          case nir_intrinsic_deref_atomic_fmin:
          case nir_intrinsic_deref_atomic_fmax:
          case nir_intrinsic_deref_atomic_fcomp_swap:
-            try_lower_direct_buffer_intrinsic(intrin, state);
+            try_lower_direct_buffer_intrinsic(intrin, true, state);
             break;
 
          case nir_intrinsic_get_buffer_size: {
diff --git a/src/intel/vulkan/anv_pipeline.c b/src/intel/vulkan/anv_pipeline.c
index b0ed2187376..09abf4e85fd 100644
--- a/src/intel/vulkan/anv_pipeline.c
+++ b/src/intel/vulkan/anv_pipeline.c
@@ -147,6 +147,7 @@ anv_shader_compile_to_nir(struct anv_device *device,
          .int8 = pdevice->info.gen >= 8,
          .int16 = pdevice->info.gen >= 8,
          .int64 = pdevice->info.gen >= 8,
+         .int64_atomics = pdevice->info.gen >= 9 && pdevice->use_softpin,
          .min_lod = true,
          .multiview = true,
          .physical_storage_buffer_address = pdevice->has_a64_buffer_access,